Contents of
October Issue

Obituary - Seymour Cray, Father of supercomputing

London, 9-10 -1996 Seymour Cray, died on October 5th, at the age of 71 after a car crash two weeks earlier. He was one of the most single minded computer engineers of his time. He is rightfully recognized as the father of supercomputing and his computer creations have been the most influential engines for change in science and technology in the last forty years.

The computers he designed and built were used as the work horse in many fields: by "Spooks" involved in encryption or cracking spy codes, by "Nukes" simulating the design of nuclear weapons, by meteorologists predicting the weather, by aeroplane and car designers and also by every other civilian scientific and engineering endeavour.

He was no armchair visionary but an active frontiers redneck of a man, who's work was at the core of actions which changed the world. Seymour was reported telling science magazine that: the fact that Los Alamos could tests H-bombs on computers was the key to the 1963 Nuclear Test Ban Treaty between the USA and the Soviet Union. Another key factor was the ability to verify seismic tremor from earthquakes and underground testing which was developed at Aldermaston in the UK.

The man who built computers

Seymour Cray was born in 1925 and brought up in Chippewa falls, Wisconsin. He graduated from the University of Minnesota in electrical engineering and also received a masters degree in applied mathematics, in 1951. After graduation, he joined the St. Pauls based Engineering Research Associates, a start up company developing cryptographic equipment for the US navy. At ERA, Cray had his chance to design his first computer, the 1103. ERA was taken over first by Remington Rand and later by Sperry, which produced one of the first commercial computer, the Univac.

In1957, Seymour Cray was one of a small team of engineers who left Sperry and started Control Data Corporation (CDC). The team, under the leadership of James Thornton, set up in a disused warehouse at 501 Park Avenue, where they designed and produced their first machine, the Control Data 1604. This was arrived at from 1103+501 =1604. The apocryphal story is that they funded the operation by selling stock at street corners. In reality the industry was picking up and venture capital was pouring in. This machine was one of the first to use Germanium transistors instead of valves or vacuum tubes.

The CDC1604 was a success and CDC expanded by seting up an engineering development laboratory at Chippewa with Thornton its first director. Seymour Cray made his name at Chippewa by designing the CDC 6600. This was built using silicon planar transistors which once invented it changed the future of computers forever. It was orders of magnitude more reliable, more tolerant to higher temperatures and it allowed higher device speeds. The CDC 6600 consisted of 400 thousand transistors, but by using silicon the system mean free time between failures due to the transistor was over 2000 hours. This was an enormous reliability inprovement compared to previous machines.

The central premise of the design of the CDC 6600 was functional parallelism. A number of independent functional units allowed the machine to do housekeeping and processing in parallel. This machine and its successors made Control Data the main provider of systems for the scientific and engineering business. By 1972, Cray was having policy conflicts with Thornton and the CDC executives on the direction of CDC business which was reflected in the architecture of the next machine. Two architectures were competing, the Cyber205 architecture and the Cray-1. This was resolved when in 1972 , Cray left Control Data and found Cray Research Inc.

The man who built supercomputers

With a $5 million budget and the use of Chippewa laboratory he designed the Cray-1, the first vector supercomputer. He delivered it to the US Nuclear Research Laboratory at Los Alamos in the Autumn of 1975. For the next year this machine was debugged and populated with software by the Los Alamos staff. This machine was running at the amazing speed of 120 Mhz; i.e. it had an 8.5 nanosecond clock. This speed was untenable by the technology of the time and I remember Seymour telling us, with tongue in cheek, that the random errors observed were caused by Cosmic Rays. The Cray-1 clock had to be slowed down to 80 Mhz before the machine could function properly. By 1976, vector processing was born.

In 1975, Cray identified a market of about 80 machines of this kind. In the event Cray Research sold over 2000 systems in 20 years, with multiple sales to the same site. Before the collapse of the Soviet Union, Western defence ministries and their associated military industrial complexes were the mainstay of Cray Research business; Los Alamos and Lawrence Livermore Laboratories owing 10 or more machines at any given time.

Habits of a life time are hard to break

The Cray-1 was followed with the design of the Cray-2, one of the most aesthetically pleasing computer of all time. This machine failed to take off in the market, but Cray had high hopes in his next creation, the Cray-3, which was to use Gallium Arsenide. GA has different electrical properties to silicon and provides an order of magnitude better device performance. The Cray-3 was using innovative material at the limit of existing technology. Special robotic tooling had to be developed and a whole new set of problems overcome.

I remember going to Chippewa in May 1988 and during our discussions I asked permission to describe the Cray-3 in my book "Supercomputers and Their Use", which I was then revising. Seymour was reluctant at first, but at the same time he was adamant that the machine will be shipped within six months. After promising not to publish before that date, permission was granted for a broad description of the Cray-3, in the new edition. It was clear to me then that this technology was not ready for use, and Cray soon got into financial straights because of escalating development costs.

In 1989, Seymour Cray was once again in conflict with policy makers and financial executives over the development of his next machine, so he left Cray Research and found Cray Computers Corporation. He moved to Colorado Springs to pursue his dream of building the Cray-3 using Gallium Arsenide. Only a quadrant of the Cray-3 was ever finished and shipped to Lawrence Livermore Laboratories. The production was bedeviled with many problems and Livermore eventually cancelled the order. This was the last straw for Cray Computers which, having spend a lot of venture capital eventually went into liquidation in 1995.

With his enormous prestige and contacts in high places, Cray started a new company SRC to pursue his single minded lifelong ambition, for always designing and producing the most powerful computer in the world. This could only be achieved by using high risk latest technology and extending the engineering frontiers.

Seymour was not a flamboyant man, but had an iron will and never suffered fools gladly. He was often charming and his a low key humour betrayed an agile mind in search for new ideas. He avoided the limelight and over exposure to marketing chi- chat which he considered a waste of his time. He often joked that he was an overpaid plumber because most of the computer design revolved around designing cooling units to extract heat from the hardware.

He also thought deeply about software. In 1976, in one of his visits to the University of London Computer Centre to promote the Cray-1, we discussed the need for a mechanism in Fortran, which allowed a user to tell the compiler that certain structures were not data dependent so that vectorised code can be produced. This problem was belatedly addressed by High Performance Fortran (HPF), almost two decades later. This is typical of most of Seymour's work, about twenty years ahead of the rest of us.

He was highly competitive and determined to get his way as some of the talented engineers, Neil Lincoln at Control Data and Steve Chen at Cray Research soon found out. When the odds went against him he just packed his bags, founded a new company and did it his way.

End of an Era - A leading light extinguished

With the advent of commodity chips and the end of the cold war, supercomputers made from custom built chips and costing $30 million are less in demand. In a sense time has taken its toll and the computer world is transforming itself by changing the subject. Even high performance MPPs made from commodity chips are having a hard time to remain on the market. Some would say that with the death of Seymour Cray, the end of the supercomputer Era is nigh.

Seymour Cray was a special kind of frontiers man with a single minded mission and the will to succeed. His world class work not only helped to shape the world, but was also an inspiration to many younger engineers. Chippewa honoured its famous son by naming a street after him. Seymour, your life's work is done, now you can rest in peace.

Seymour Cray, world class supercomputer designer and engineer, born September 28, 1925; died October 5, 1996.

Chris Lazou

Top of Article

© The HOISe-NM Consortium 1996